Embedded Files
the use of validated questionnaires. We used a purposely-built form to extract data. We developed a pilot version which was modified after assessment of the first 10 studies. Any disagreement between authors was solved by consensus and, if not reached, by discussion with a third author (BSP). Data missing from the primary studies was requested by contacting the authors. Duplicate or overlapping reports were identified, privileging the ones with more complete results. Quality assessment An adapted version of The National Health Institute (NIH) Quality Assessment Tool for Observational Cohort and Cross-Sectional Studies was used to evaluate the included primary studies(15). The tool consists of 14 items with “Yes/No” answers. Question number 8, inquiring if the study examines different exposure levels and their relation with the outcome, was not answered for any study due to the dichotomous format of the data retrieved. For cross-sectional studies, no answers were registered for questions 6, 7, 10, 12 and 13, as such questions would not be applicable. The two authors (JR and FFP) independently assessed the quality of each study, with any disagreement being solved by consensus. Data analysis As our main analysis, we performed meta-analysis assessing the associations between AR and anxiety and depression, based on adjusted odds ratios (ORs) (i.e., OR obtained with multivariable regression models) retrieved from included primary studies. Whenever primary studies applied more than one multivariable model, we retrieved the results obtained with the model adjusting for the largest number of independent variables. In order to assess results consistency, we also performed meta-analysis for the crude associations between AR and anxiety/depression, by pooling raw data retrieved from included primary studies. For those studies only providing effect size measures and the respective confidence intervals, raw data was estimated using the methods described by Di Pietrantonj and using estimraw interactive tool(16,17). Pooled ORs were estimated by means of random-effects metaanalysis, with weighting by the restricted maximum likelihood approach. In each meta-analysis, we only included primary studies with the same design (i.e., we did not include crosssectional and cohort studies in the same analyses). Betweenstudy heterogeneity was assessed using the I2 statistic and the Cochran Q statistic.
Report abuse